Quadrupedal robots have played a crucial role in various environments, from structured environments to complex harsh terrains, thanks to their agile locomotion ability. However, these robots can easily lose their locomotion functionality if damaged by external accidents or internal malfunctions. In this paper, we propose a novel deep reinforcement learning framework to enable a quadrupedal robot to walk with impaired joints. The proposed framework consists of three components: 1) a random joint masking strategy for simulating impaired joint scenarios, 2) a joint state estimator to predict an implicit status of current joint condition based on past observation history, and 3) progressive curriculum learning to allow a single network to conduct both normal gait and various joint-impaired gaits. We verify that our framework enables the Unitree’s Go1 robot to walk under various impaired joint conditions in real-world indoor and outdoor environments.

In the walking control of a biped humanoid robot, the walking performance is seriously affected by the initial home posture. An important thing is that the initial home posture of real robot is slightly different at every setting because a zero position of joint is not exactly the same. Accurate and consistent initial home posture is essential when we compare and analyze walking control algorithms. In general, the conventional method for electric motor uses an incremental encoder with a limit switch or an absolute encoder such as potentiometer to make a zero position. However, this method for the multi-axis humanoid robot is difficult, time-consuming, and inaccurate. Furthermore, it has the disadvantage that additional limit switch or absolute encoder can interfere with the design objective of compactness. This paper describes a novel adjustment method of home posture for a humanoid robot utilizing incremental encoders, an inertial sensor and force torque sensors. Four kinds of controllers are proposed and adjusted offsets are measured when outputs of the controllers are converged. The experimental results from KHR-2 show the effectiveness of the proposed algorithm.

This paper describes an algorithm about online gait trajectory generation method, controller for walking, brief introduction of humanoid robot platform KHR-3 (KAIST Humanoid Robot-3: HUBO) and experimental result. The gait trajectory has continuity, smoothness in varying walking period and stride, and it has simple mathematical form which can be implemented easily. It is tested on the robot with some control algorithms. The gait trajectory algorithm is composed of two kinds of function trajectory. The first one is cycloid function, which is used for ankle position in Cartesian coordinate space. Because this profile is made by superposition of linear and sinusoidal function, it has a property of slow start, fast moving, and slow stop. This characteristics can reduce the over burden at instantaneous high speed motion of the actuator. The second one is 3rdorder polynomial function. It is continuous in the defined time interval, easy to use when the boundary condition is well defined, and has standard values of coefficients when the time scale is normalized. Position and velocity values are used for its boundary condition. Controllers mainly use F/T(Force/Torque) sensor at the ankle of the robot as a sensor data, and modify the input position profiles (in joint angle space and Cartesian coordinate space). They are to reduce unexpected external forces such as landing shock, and vibration induced by compliances of the sensors and reduction gears, because they can affect seriously on the walking stability. This trajectory and control algorithm is now on the implementing stage for the free-walking realization of KHR-3. As a first stage of realization, we realized the marking time and forward walking algorithm with variable frequency and stride

To celebrate the 100th anniversary of the announcement of the special relativity theory of Albert Einstein, KAIST HUBO team and hanson robotics team developed android type humanoid robot Albert HUBO which may be the world's first expressive human face on a walking biped robot. The Albert HUBO adopts the techniques of the HUBO design for Albert HUBO body and the techniques of hanson robotics for Albert HUBO's head. Its height and weight are 137cm and 57Kg. Albert HUBO has 66 DOFs(31 for head motions and 35 for body motions) And head part uses 'Fubber' materials for smooth artificial skin and 28 servo motors for face movements and 3 servo motors for neck movements are used for generating a full range of facial expressions such as laugh, sadness, angry, surprised, etc. and body is modified with HUBO(KHR-3) introduced in 2004 to join with Albert HUBO's head and 35 DC motors are embedded for imitating various human-like body motions.

In this paper, we describe the mechanical design, system integration and dynamic walking of the humanoid, KHR-2 (KAIST Humanoid Robot– 2). KHR-2 has 41 DOFs in total, that allows it to imitate various human-like motions. To control all joint axes effectively, the distributed control architecture is used, which reduces computation burden on the main controller, and allows convenient system. A servo motor controller was used as the sub-controller, whereas a 3-axis force/torque sensor and an inertia sensor were used in the sensory system. The main controller attached on the back of KHR-2 communicates with the sub-controllers in real-time through CAN (Controller Area Network) protocol. Windows XP was used as the operation system, whereas RTX HAL extension commercial software was used to realize the real-time control capability in Windows environment. We define the walking pattern and describe several online controllers in each stage. Some of the experimental results of KHR-2 are also presented.